Mercor pays out more than $2 million per day to 30,000-plus weekly active contractors — the largest credentialed RLHF workforce in the United States. Its rate architecture, a four-tier structure from $12/hr to $200-plus, is the reference pricing grid for every US-based RLHF worker trying to answer what their expertise is actually worth.
The Rate Card, Tier by Tier
Mercor's public rate data, aggregated across HireFeed, aitraining.jobs, and the company's own resources page, describes a consistent structure as of Q3 2026.
Tier 1: $12–$25/hr. Generalist text evaluation, basic preference annotation, translation review. This is the entry tier — accessible to workers without domain credentials, primarily used for volume annotation tasks that do not require subject-matter depth. This is also where Mercor overlaps most directly with Outlier and DataAnnotation in terms of task type, though Mercor's floor ($12/hr) sits higher than Outlier's effective floor ($10–$14/hr for the lowest-complexity tasks).
Tier 2: $25–$55/hr. Specialized RLHF preference annotation, coding evaluation, multilingual evaluation in high-demand languages. A software engineer with two-plus years of production experience can realistically enter at $35–$40/hr here. The median Tier 2 worker earns $40–$55/hr in Q3 2026, per platform rate aggregates. This tier saw the most new supply come online in 2026 as computer science graduates recognized that Mercor's vetting bar, while real, is passable with a credible engineering background and a strong benchmark performance.
Tier 3: $55–$130/hr. Domain expert evaluation — advanced engineering, graduate-level math and science, financial modeling review, legal reasoning evaluation. The p50 for Tier 3 domain-expert RLHF work sits at $130/hr in Q3 2026, per ENTRA Talent Index data. A PhD in a STEM field who can pass Mercor's benchmarking round and demonstrate consistent calibration against expert raters enters this tier by default. An attorney evaluating legal AI outputs, a CFA-holder reviewing financial reasoning, or a biomedical researcher auditing medical AI responses all operate in Tier 3.
Tier 4: $130–$200+/hr. Constitutional AI specialists, alignment feedback designers, RL environment architects. Board-certified physicians reviewing clinical AI outputs earn $90–$160/hr depending on specialty. The deepest-credentialed specialists at the p90 earn $220/hr, per ENTRA Q3 2026 platform data. For context: a Tier 4 specialist working the equivalent of full-time hours annualizes at $270K–$460K before Mercor's accuracy bonuses for high-calibration raters.
The blended platform average — pulled upward by the Tier 3 and 4 concentration — sits at $85/hr. That figure, which Mercor co-founder and CEO Brendan Foody announced via LinkedIn in late 2025, is the number that gets cited when the platform markets itself to potential contractors. What it obscures is the distribution: a majority of active contractors earn $25–$55/hr, with the average elevated by a smaller population of high-credentialed specialists doing Tier 3 and 4 work.
Why the Average Hides the Structure
The $85/hr average is not a misrepresentation. It is a compression. The platform's economics depend on a contractor mix that looks nothing like a traditional staffing firm. Mercor operates with an estimated 300 internal employees against a contractor base of 300,000-plus registered professionals, with 30,000-plus active on a weekly basis. The company pays out 60–70% of topline revenue to contractors, meaning its net revenue on a $2B gross run rate is roughly $600M–$800M — a margin profile that competes with premium software businesses, not staffing agencies.
The rate architecture is designed to signal quality to lab clients, not to maximize contractor take-home across all tiers. OpenAI, Anthropic, and Google DeepMind collectively account for more than 90% of Mercor's revenue, per public reporting. Those three clients need credentialed-professional RLHF at scale — not commodity annotation. The platform rate card serves as a self-selection mechanism: workers willing to go through Mercor's vetting (identity verification, credential screening, and a multi-round benchmark evaluation) earn more than they would anywhere else. Workers who want easier onboarding use Outlier.
The Competitive Collapse: What Happened to Outlier
Six months ago, Outlier — Scale AI's consumer-facing contractor platform — was the volume competitor to Mercor's credentialed model. Today it is not a credible alternative for the work that matters.
The sequence: Meta invested $14.3 billion for a 49% stake in Scale AI in June 2025 and installed Scale's CEO inside Meta's AI organization. That transaction immediately dried up Scale's non-Meta client roster. OpenAI terminated its Scale relationship. Google pulled back. Scale cut roughly 200 employees and 500 contractors in the weeks following the deal. Outlier, which is where Scale's contractor workload flows, experienced the client attrition directly: queues emptied. Pay compression followed — from $40–$50/hr for technical tasks in mid-2025 to $18–$25/hr by early 2026, per contractor community reporting on paidtotrainai.com and aiworkfinder.com.
The Outlier problem is structural, not cyclical. The platform's task volume is a downstream function of Scale AI's lab contracts, and Scale's lab contract base is now concentrated in a single customer that happens to be its majority owner. An Outlier contractor working in Q4 2026 is working for Meta's AI training needs, mediated through Scale. The platform diversity that made Outlier attractive in 2024 — projects from OpenAI, Google, and Anthropic all flowing through the same contractor pool — does not exist anymore.
For the US-based AI trainer looking to build a sustainable income stream from RLHF work, the competitive landscape has effectively resolved: Mercor for credentialed professionals, DataAnnotation for mid-tier generalists, Outlier for opportunistic fill-in work when queues are not empty. The platform tier list that was genuinely competitive in 2024 is now stratified.
Platform Economics and the Concentration Risk
Mercor's $2B gross annualized run rate — confirmed as of June 2026, up from $760M at end-2025 — reflects three years of compounding client concentration. More than 90% of revenue from three customers is not a balanced marketplace; it is a B2B services business with a consumer-facing interface. The company priced its Series C at a $10B valuation in October 2025, led by Felicis with Benchmark and General Catalyst participating. By July 2026, it was in discussions for a $20B valuation — a doubling in nine months on the back of the revenue trajectory.
Foody's thesis, stated most clearly at a Stanford ETL talk in April 2026, is that Mercor is infrastructure for the agentic AI economy: "Everyone's been focused on what models can do. But the real opportunity is teaching them what only humans know — judgment, nuance, and taste." That framing positions the platform not as a labor marketplace that might be automated away, but as the data supply chain for the next layer of model capability. Whether that thesis survives a material shift toward synthetic data in post-training workflows is the open empirical question. It is also the question every RLHF trainer on the platform is implicitly making a bet on.
Three Things the October Rate Data Tells You
The $55/hr threshold is the practical credential floor. Below it, supply is abundant and rates are competitive with offshore annotation economies. Above it, the credential-verification layer creates meaningful supply constraint. A contractor who can clear the Tier 2-to-Tier 3 transition — through graduate credentials, strong benchmark performance, or demonstrated specialization in a domain the labs are actively paying for — moves into a rate band where demand reliably exceeds supply.
Coding and math evaluation are the fastest path to Tier 3. The labs pay $60–$80/hr for rigorous technical problem evaluation — tasks that require a practitioner who can identify not just that a code solution is wrong, but why, and write the correction in a form a training engineer can use. A software engineer with strong CS fundamentals can qualify for this work without an advanced degree. No other route to the $60–$80/hr range on the platform is as accessible.
The platform premium over Outlier is now structural, not marginal. In 2024, the choice between Mercor and Outlier involved meaningful tradeoffs — easier onboarding on Outlier, higher ceiling on Mercor, similar mid-range rates. In October 2026, Outlier's effective mid-range rate for technical work is $18–$30/hr. Mercor's Tier 2 floor for comparable work is $35–$40/hr. The credential bar is the entry cost. For workers who can clear it, the arbitrage is no longer close.
The RLHF labor market that Mercor built is not the gig-economy annotation market that Scale AI was selling in 2022. It is a credentialed-professional services market that happens to route through a platform interface. The $85/hr average is the number. The rate architecture behind it is the story.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.